Meta Superintelligence Labs has unveiled Muse Spark 1.1, a significant advancement in multimodal reasoning models designed for agentic tasks. Released on July 9, 2026, alongside a public preview of the Meta Model API, this model marks a major step forward in the company’s efforts to build AI systems capable of complex, autonomous decision-making.
Key Features of Muse Spark 1.1
The new model boasts a 1,000,000-token context window, which it actively compacts to optimize performance. This feature allows Muse Spark 1.1 to process and reason over exceptionally long inputs, a critical capability for tasks requiring deep contextual understanding. Additionally, the model supports zero-shot generalization to new tools and MCP (Model Control Protocol) servers, enabling it to adapt quickly to unfamiliar environments without prior training.
One of its standout capabilities is multi-agent delegation, where the model can manage and coordinate parallel subagents to execute complex workflows. This functionality is particularly valuable for agentic tasks that require collaboration between multiple AI components, enhancing both scalability and task execution efficiency.
Performance and Positioning
According to Meta’s internal benchmarking, Muse Spark 1.1 leads in tool usage capabilities, demonstrating strong performance in interactive and dynamic environments. However, it trails behind models like Opus 4.8 and GPT-5.5 in coding benchmarks. This reflects the ongoing challenges in balancing multimodal reasoning with specialized tasks like software development.
The launch of Muse Spark 1.1 and the Meta Model API underscores Meta’s ambition to become a leading platform for advanced AI agents. As the industry moves toward more autonomous and intelligent systems, models like Muse Spark 1.1 are pivotal in shaping the future of AI-driven automation.
